Improve Resource Management and Update Dependencies (#24)

* clear connections

* tritony==0.0.20

.github/workflows/pre-commit_pytest.yml

@@ -42,17 +42,18 @@ jobs:
         --shm-size=1g
         --user=root
     steps:
-      - uses: actions/checkout@v3
+      - uses: actions/checkout@v4
+      - uses: astral-sh/setup-uv@v5
       - name: test bash
         env:
           TRITON_HOST: localhost
           TRITON_HTTP: 8000
           TRITON_GRPC: 8001
           TRITON_RETRIES: 2
           TRITON_CLIENT_TIMEOUT: 1
+          UV_SYSTEM_PYTHON: 1
         run: |
           tritonserver --model-repo=$GITHUB_WORKSPACE/model_repository &
-          pip install uv
           uv pip install .[tests]
           sleep 3
 

tests/test_model_call.py

@@ -59,13 +59,3 @@ def test_with_optional(config):
         }
     )
     assert np.isclose(result[0], sample[0] - OPTIONAL_SUB_VALUE, rtol=EPSILON).all()
-
-
-def test_reload_model_spec(config):
-    client = get_client(*config, model_name="sample_autobatching")
-    # force to change max_batch_size
-    client.default_model_spec.max_batch_size = 4
-
-    sample = np.random.rand(8, 100).astype(np.float32)
-    result = client(sample)
-    assert np.isclose(result, sample).all()

tritony/tools.py

@@ -318,11 +318,18 @@ async def async_default_model_spec(self):
             await self._async_renew_triton_client(self._triton_client)
         return self.model_specs[self.default_model]
 
+    async def async_close(self):
+        await self._triton_client.close()
+
     def __del__(self):
-        # Not supporting streaming
-        # if self.flag.protocol is TritonProtocol.grpc and self.flag.streaming and hasattr(self, "triton_client"):
-        #     self.triton_client.stop_stream()
-        pass
+        try:
+            if asyncio.iscoroutinefunction(self._triton_client.close):
+                if not asyncio.get_event_loop().is_closed():
+                    asyncio.create_task(self._triton_client.close())
+            else:
+                self.close()
+        except Exception:
+            pass
 
     @retry((InferenceServerException, grpc.RpcError), tries=TRITON_RETRIES, delay=TRITON_LOAD_DELAY, backoff=2)
     def _renew_triton_client(

tritony/version.py

@@ -1 +1 @@
-__version__ = "0.0.19"
+__version__ = "0.0.20"